Results are in: static foreach is a slower than hand unrolling your

[bearophile]

You can try LDC compiler too (D1 only) with few changes.
Note: LDC isn't using LLVM 2.7 yet.

Bye,
bearophile